Accelerating the Exploration of Top Interface Passivators via Machine Learning for High‐Performance Perovskite Solar Cells

Why It Matters

The framework cuts months of trial‑and‑error, accelerating commercial‑grade perovskite solar cell development and lowering R&D costs.

Key Takeaways

• ML screened 162 ammonium salts for perovskite passivation
• XGBoost excelled for n‑i‑p architecture predictions
• RDKit SVR best for p‑i‑n devices
• SHAP highlighted concentration, weight, polarity, I/Br ratio
• Experimental validation achieved 1.58% error

Pulse Analysis

Machine learning is reshaping materials discovery by turning vast chemical spaces into tractable design problems. In the perovskite solar cell arena, the new workflow leverages curated literature data and multiple molecular fingerprints to train architecture‑aware regression models. By separating n‑i‑p and p‑i‑n configurations, the researchers achieved sub‑0.05 mean absolute error, demonstrating that tailored algorithms outperform one‑size‑fits‑all approaches. This precision enables rapid virtual screening, narrowing thousands of candidates to a handful of high‑impact passivators without costly synthesis cycles.

Interpretability tools such as SHAP further differentiate this study from typical black‑box models. By quantifying the influence of passivator concentration, molecular weight (120‑280 g mol⁻¹), surface polarity fragments, and the iodide‑bromide ratio, the analysis provides actionable design rules for chemists. These insights bridge the gap between data science and synthetic chemistry, allowing researchers to engineer molecules that directly target the most influential parameters for defect mitigation and charge extraction.

The practical payoff is evident: the top‑ranked ammonium salt boosted the power‑conversion efficiency of a Cs₀.₀₅FA₀.₉₅PbI₃ p‑i‑n device by 15.2%, matching the model’s prediction within 1.6% error. Such alignment validates the predictive power of the workflow and signals a path toward scalable, high‑efficiency perovskite modules. As the solar market seeks low‑cost, high‑performance alternatives to silicon, these data‑driven methods could shorten time‑to‑market, reduce R&D expenditure, and accelerate the adoption of next‑generation photovoltaic technologies.